Entropy Data Sync

Step 4 — Apply the fixes

For each missing artifact, copy the corresponding template from ${PLUGIN_ROOT}/skills/entropy-data-sync/templates/ into the user's project, substituting placeholders. Do not overwrite existing files; if a file is present but incomplete, surface the diff and ask before changing.

When (and only when) you're about to write openlineage.yml or .github/workflows/data-product.yml, resolve API_HOST from the active CLI connection:

entropy-data connection get -o json

Use the host field to substitute {{API_HOST}} in those templates. Self-hosted deployments are handled via entropy-data connection add --host <host>, not a plugin-level setting.

If the ODPS file exists but was flagged as incomplete — missing dataProductBuilder customProperty in Step 2, append the entry to the top-level customProperties list (do not reorder or touch other entries):

customProperties:
  - property: "dataProductBuilder"
    value: "https://github.com/entropy-data/dataproduct-builder-dbt"

Surface the diff and ask before saving.

The templates live at:

• templates/data-product.odps.yaml → write to <DATA_PRODUCT_ID>.odps.yaml
• templates/models/output_ports/v1/contract.odcs.yaml → write to <CONTRACT_PATH> (i.e. models/output_ports/v1/<CONTRACT_FILE>)
• templates/openlineage.yml → write to openlineage.yml
• templates/.github/workflows/data-product.yml → write to .github/workflows/data-product.yml

If the audit reported a legacy datacontracts/ directory, ask the user before moving its contents. The default migration is:

• datacontracts/<contract>.odcs.yaml → models/output_ports/v1/<contract>.odcs.yaml (if it matches an output port). If multiple output port versions exist, ask which one.
• Update the git connection's --repository-path for each moved contract (Step 4b).
• Delete the now-empty datacontracts/ directory only after the user confirms the move.

This skill does not create input-port ODCS files. They appear only when dataproduct-implement resolves an access agreement and writes the cached upstream contract to models/input_ports/. If the audit found stale input-port ODCS files (no matching .source.yaml), surface them as orphans and let the user decide whether to delete them.

For the model layout, create the directories models/input_ports/, models/staging/, models/intermediate/, models/output_ports/v1/ if absent, plus _models.yml placeholders so dbt does not warn about empty directories. Do not move existing models — only add the empty subfolders the user is missing, and note it in the report.

Also update dbt_project.yml's models: block so the materializations match the reference (output port = table, staging/intermediate = view):

models:
  <DBT_PROJECT_NAME>:
    +materialized: table
    staging:
      +materialized: view
    intermediate:
      +materialized: view

If the models: block already exists, only add missing keys; do not clobber the user's customizations.

Step 4b — Configure git connections

Only run this sub-step if the audit (Step 2) flagged at least one git connection as missing or the user confirmed re-creating a drifted one. Skip entirely if every connection is already correct, if the audit marked them as deferred, or if the working directory is not a git repository (check with git rev-parse --is-inside-work-tree — if it errors or returns false, there's no remote to register; tell the user to run git init and add a remote first, then re-run this skill).

For the data product:

entropy-data dataproducts gitconnection put <DATA_PRODUCT_ID> \
  --repository-url <GIT_REPOSITORY_URL> \
  --repository-path <ODPS_FILE> \
  --repository-branch <GIT_REPOSITORY_BRANCH> \
  --git-connection-type <GIT_CONNECTION_TYPE> \
  [--host <GIT_HOST>] \
  [--git-credential-external-id <GIT_CREDENTIAL_EXTERNAL_ID>]

For each output-port ODCS file (models/output_ports/**/*.odcs.yaml):

entropy-data datacontracts gitconnection put <CONTRACT_ID> \
  --repository-url <GIT_REPOSITORY_URL> \
  --repository-path <CONTRACT_PATH> \
  --repository-branch <GIT_REPOSITORY_BRANCH> \
  --git-connection-type <GIT_CONNECTION_TYPE> \
  [--host <GIT_HOST>] \
  [--git-credential-external-id <GIT_CREDENTIAL_EXTERNAL_ID>]

Do not register git connections for input-port ODCS files. They are cached copies of upstream contracts; the upstream data product owns the canonical record.

Notes:

• --repository-path is relative to the repo root, not the working directory. The ODPS path is just <DATA_PRODUCT_ID>.odps.yaml; output-port contract paths look like models/output_ports/v<N>/<CONTRACT_FILE>.
• Omit --host for SaaS providers (github.com, gitlab.com, bitbucket.org, dev.azure.com); set it only for self-hosted instances.
• These commands fail if the underlying data product / contract does not exist on the platform yet. If you skipped earlier because of "deferred," surface the manual command in Step 5 so the user can run it after the first workflow run. Do not retry-loop.
• If the audit reported drift (existing connection with different URL/branch/path), confirm with the user before overwriting — put is upsert.

Step 5 — Final report

Always end with this exact two-part format so the user gets a consistent recap.

Part 1 — outcome table. One row per artifact from the audit. Use the Status enum below; Details is a short, plain-text note (file path, or "—" if nothing to add).

Status enum (use exactly these words):

• created — the skill wrote a new file or registered a new connection.
• updated — the skill patched an existing file or fixed a drifted connection.
• already present — no change needed.
• deferred — skipped intentionally (data product/contract not yet published, or no git repo). The deferred command(s) appear in Part 2.
• skipped — the user declined when asked to confirm.

Part 2 — next steps. Bullet list, only include the items that apply:

• For each deferred git connection, the exact entropy-data dataproducts gitconnection put … or entropy-data datacontracts gitconnection put … command to run after the first CI publish.
• "Set GitHub repository secrets: ENTROPY_DATA_API_KEY, plus platform creds (DBT_DATABRICKS_HOST, DBT_DATABRICKS_HTTP_PATH, DBT_DATABRICKS_TOKEN for Databricks; equivalents for other platforms)."
• "Fill in the data contract schema in <CONTRACT_PATH> — the template only seeds id and updated_at."
• "Run dbt-ol run locally once to verify lineage flows to Entropy Data (requires OPENLINEAGE__TRANSPORT__AUTH__APIKEY)."

If there is nothing in Part 2, write a single line: No further action required.

Conventions and constraints

• Platform-aware workflow: the workflow template assumes Databricks. If the user picks Snowflake/BigQuery/Postgres, swap the dbt-databricks install line, the Create profiles.yml block, and the DATACONTRACT_* env vars to the matching dialect. Do not generate a Databricks workflow for a non-Databricks project.
• No invented schema: when generating the ODCS file, do not invent columns. Seed it with id + updated_at and tell the user to fill in the rest, or — if dbt models already exist for the output port — derive columns from _models.yml if available.
• Idempotent: running the skill a second time should be a no-op when everything is already present. For git connections that means: if gitconnection get returns a record matching the local repo URL / branch / path, do not call put.
• Don't overwrite drifted git connections silently. If the platform reports a different URL/branch/path than the local repo, surface the diff and ask. The user may have a fork, a renamed default branch, or a deliberate path remap.
• Don't push secrets: never write API keys, tokens, or hostnames into committed files. They must come from GitHub secrets in the workflow.
• Don't create a git repo or commit: leave VCS state to the user.